We start with the tasks inside a public occupation, estimate how AI may support each task, attach external evidence only when the match is explicit, and roll the task results into a broad role range. The path remains visible so the range is never the only thing you can inspect.
Role-level answer
The range includes work where AI can assist with preparation, drafting, analysis, review, information handling, or execution. A task can change substantially while a person still owns the decision and outcome.
Under 30%
AI may affect a smaller share of the analyzed task set.
30–45%
AI may change a meaningful share of preparation and execution work.
45%+
AI may reach many tasks, while human responsibility can still remain.
Task-level process
01
Use the complete captured public task set for the occupation.
02
Look at AI capability, task structure, and where human involvement still matters.
03
Attach public task evidence only through a direct identifier or explicit wording match.
04
Show a role range, task bands, plain-language reasoning, and inspectable sources.
Evidence rules
Strongest task match
The external evidence points to the same occupation and task identifier. It may inform that task's work-change view.
Supporting observed-use match
Observed AI use appears against equivalent task wording. It is supporting context, not proof that the whole task is automatable.
Comparison only
The evidence describes an occupation or broad work activity. It stays visible as context and is not assigned to individual tasks.
Missing observed-use evidence is treated as missing coverage—not as a zero-impact estimate. See the source registry for dataset-specific join rules and caveats.
Limitations
Inspect further
The source registry lists datasets and retrieval versions. The Task Graph overview shows how a role, task, estimate, evidence row, and source connect in the product.